Research On Large-scale Multi-objective Unit Commitment Based On Vector Ordinal Optimization Method

The power system which operates safely, reliable, economic and superior, has been impacted significantly by reasonable unit commitment. With the development of society, objective people pursued turned to diversity greatly, in the past, minimizing the system operation cost had been focused on by unit commitment, now, economic, environment and other benefits has been concerned. In response to national policies, accelerating the implementation of generating power cleanly, so as to energy-saving power generation dispatching, in this paper, coal consumption, cost of purchasing power and emissions of SO2 are selected as objectives, feasible multi-objective unit commitment model has been built, and then solved.In this paper, the on/off schedule of generators, satisfying all constraints strictly, are defined as feasible designs, i.e., the discrete variables included by unit commitment problem. Unit commitment is a typical high-dimensional, non-convex, dynamic, mixed integer nonlinear programming problem. In the process of searching the optimal designs, "dimension disaster" would be encountered inevitably, from the perspective of engineering, vector ordinal optimization method(VOO) is introduced for the first time, search good enough designs meeting the engineering need directly. The search space is transformed from design space to a finite set, as the crude models, BP neural network is used to estimate, sort, layer all the designs in the set, after that, OPC curve is obtained and selected set is determined. The GAMS-CONOPT is used as accurate model, estimates all designs in the selected set, line membership is introduced to describe the accurate objective values, and the one with the maximum membership value would be selected as the compromise optimal design.Except above, stochastic wind is considered, based on the scenario method, Latin hypercube sampling method is used to simulate the actual wind output by sampling scenarios, sampling scenarios are reduced based on probability metric and fast forward reduction technology. The constraint, generators’ output adjustment in the same period time is introduced between sampling scenarios and forecasted scenario, so as to sampling scenarios and sampling scenarios, is considered, make the model built in this paper has much stronger robustness in dealing with the stochastic wind.Taking a real provincial power system as a numerical example, hydropower, nuclear power, biomass energy, gas turbines, thermal power and other complex power structures has been considered. The calculation results using VOO proposed in this paper are compared with mixed integer nonlinear programming by GAMS-BARON. It is found that the calculation speed of the former is several times faster than the latter with little deviation. Therefore, the VOO based method proposed by this paper is proven to be effective to solve large-scale multi-objective unit commitment problem.